摘要
图切是一种基于图论的图像分割方法,它基于最大流/最小割定理实现能量函数最小化,其中能量函数的设定、实现的流程等方面可以改进以提高对不同图像的适应性.本文给出了一个基于图切的交互式图像分割方法,用户通过手绘封闭或不封闭线条的方法提供关于前景和背景的先验信息,在此基础上实现图像的分割.首先采用分水岭方法对输入图像进行预分割,把颜色相近的像素分为若干个小区域;设定合适的能量函数,将预分割的区域之间的颜色相似性、分割结果的平滑性等约束包含在能量函数中,利用求最大流的方法求取使得能量函数最小化的标签集合,从而实现图像的分割.实验证明,本文的方法能快速有效地实现交互式图像分割.
Graph cut is a kind of image segmentation method based on graph theory. Graph cut realizes the minimization of energy based on max-flow / min-cut theorem. In order to make the method more suitable to various images,the energy function and the work flow of the method can be improved. This paper presents an image segmentation method based on graph cuts. In this method,users can present information about foreground and background through hand drawing closed or unclosed curves,and the image segmentation can be further realized based on the information. Watersheds method is first used to pre-segment the input image,and the image is segmented into many small regions based on the color of the pixels. An appropriate energy function is set. The energy function incorporates the color similarity between different regions and smoothness of segmentation result. The label set which makes the energy function minimized is achieved through max-flow method. Thus the image segmentation is completed. The experimental results show that this method can realize interactive image segmentation quickly and effectively.
引文
[1]Li Y,Sun J,Tang C K,et al.Lazy snapping[C]//International Conference on Computer Graphics and Interactive Techniques.New York,2004:303-308.
[2]Rother C,Kolmogorov V,Blake A.“GrabC ut”:interactive foreground extraction using iterated graph cuts[J].ACM Transactions on Graphics(TOG),2004,23(3):309-314.
[3]Wu Xiaoyu,Wang Yangsheng.Interactive foreground/background segmentation based on graph cut[C]//International Congress on Image and Signal Processing.Sanya,2008.
[4]Boykov Y,Kolmogorov V.An experimental comparison of min-cut/max-flow algorithms for energy minimization in vision[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2004,26(9):1 124-1 137.
[5]Boykov Y,Jolly Pi M.Interactive graph cuts for optimal boundary and region segmentation of objects in n-d images[J].ICCV,2001(7):105-112.
[6]Vincent L,Soille P.Watersheds in digital spaces an efficient algorithm based on immersion simulation[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,1991,13(6):585-598.
